Model selection of M-estimation models using least squares approximation

This paper proposes a new criterion for the M-estimation models based on a least squares approximation, which is proved to be selection consistent. Compared with the existing criteria, this new one has two attractive features. One is that model selection based on it has much lower computational cost. The other is that it may bring considerable improvement in some cases since it is essentially based on the efficient GMM estimation. (C) 2015 Elsevier B.V. All rights reserved.